[prakashUXtech]

Anatomy round 1, batch r1-b1 for effects-engine (#433). Fixes the blur-family findings in Sources/CoreRaster/Effects/BlurFamily.swift and adds the tests and mutation entries those contracts were missing.

Fixes

1. Box blur reach documented. boxBlurred(radius:) now says that each pass averages a 2·radius+1 px window, that colour spreads up to 3·radius px, that the result is close to a Gaussian of sigma ≈ sqrt(radius·(radius+1)), and that a caller should pad by 3·radius. Proved by BlurFamilyTests.boxReachesThreeRadii: at radius 2 a dot is nonzero 6 px away and exactly zero 7 px away, on both axes.
2. Motion blur centring documented. motionBlurred(degrees:distance:) and the file header now say the blur is centred, so only the axis matters. They also give the integer-distance snap and say that an odd integer distance sits 1/2 px toward -angle.
3. Motion blur is centred (contract fix). Before, the first sample offset was always floor(-distance/2), so a 0.4 px blur moved the layer 0.8 px. It now starts exactly at -distance/2. Exact centring for every distance broke the integer-distance acceptance test motionSpreadsTen (half-pixel samples spread a 9 px blur over 11 columns). So the code takes the snap option: an integer distance snaps its first sample down to a whole pixel (floor(-distance/2), as before), and every non-integer distance is centred exactly. Proved by BlurFamilyTests.motionIsCentred: for d in 0.4, 2.5 and 9 at 0 and 180 degrees, the centroid stays within 1/2 px of 16 in x and on row 16. For the non-integer distances, 0 and 180 degrees give the same centroid within 1e-6. Before the fix d = 0.4 gave 16.8. Mutation: "motion blur floors its first sample offset".
4. Box blur is pinned to both axes and three passes. boxKeepsEnergy now also requires the occupied rows to grow. The new BlurFamilyTests.boxSpreadsBothAxesThreePasses checks that a dot at radius 1 covers exactly rows and columns 13...19 and that its centre alpha is (7/27)^2. Mutations: "box blur skips its vertical pass" and "box blur runs one pass per axis instead of three".
5. Edge extension is tested. BlurFamilyTests.opaqueEdgesStayOpaque: an opaque red 12x12 buffer stays exactly red and opaque at every pixel, corners included, under box radius 3 and motion distance 7. Mutation: "box blur pads before the start of a line with zeros".
6. Right-angle snap is tested. horizontalSpreadsColumns now checks for exact zeros where the cos/sin residue would leak. These are pixels at the buffer edge whose off-centre sample would land just past column or row 0. The interior pixel the finding suggested, (17,16), does not detect the leak, because the residue rounds away next to 16. Mutation: "motion blur keeps the cos residue at 90 degrees".
7. The inclusive cap is checked from the accepting side. overCapThrows now also requires radius 1,000 and distance 1,000 to return a buffer. Mutation: "blur reach check excludes its upper bound".

scripts/ci: CI: PASS on the head below. scripts/mutate --run --changed-since origin/dev: 9/9 caught. No platform API is new here (only Double.rounded, already in use), so no docs lookup was needed.

Note: the requested branch name lead/effects-engine/r1-b1 cannot exist, because the branch lead/effects-engine already exists. This PR uses lead/effects-engine-r1-b1 instead.

Refs #433 (feature-lead issue; this batch does not close it)

Belt-Lead: act-1a150bfbcf7-sfwy
Feature: effects-engine
Slice: FX-blur
Mutate: 9/9 CoreRaster.BlurFamily
Fast-Check: pass
Head: cfd4407

---

[github-actions]

Fast lane: merged into dev (123 lines, 3 files, Fast-Check marker).